Decart’s new world model can simulate hours of photorealistic driving — with some caveats

Decart, an AI startup focused on generative world models, has unveiled Oasis 3, a real-time system capable of generating photorealistic driving environments for extended periods. The new world model represents a significant advancement in autonomous vehicle (AV) testing infrastructure, offering developers an API-accessible tool for simulating complex driving scenarios without relying on pre-recorded footage or static environments.

Oasis 3 marks a departure from traditional simulation approaches by generating dynamic, photorealistic environments in real-time. The system can sustain hour-long simulations of driving scenarios, providing autonomous vehicle developers with extended sequences for training and testing perception systems. Unlike previous generative models limited to short video clips, Oasis 3 maintains visual consistency and physical plausibility across extended timeframes. The technology is now available through an API, enabling external developers and organizations to integrate the world model into their AV development pipelines.

The system's real-time generation capabilities address a longstanding challenge in autonomous vehicle testing: creating diverse, realistic, and scalable test environments without the expense and limitations of physical test tracks or extensive video dataset collection.

• Accelerated AV Development: Extended photorealistic simulations reduce dependency on costly real-world testing and data collection
• Scalable Testing Infrastructure: Developers gain access to virtually unlimited scenario variations without geographic constraints
• Democratized Access: API availability enables smaller companies and researchers to conduct sophisticated AV testing previously requiring substantial resources
• Data Generation: The system can produce diverse training data for perception models across various weather conditions, times of day, and traffic scenarios
• Reduced Development Costs: Simulation-based testing decreases expenses associated with physical prototyping and fleet testing operations

Oasis 3 addresses critical bottlenecks in autonomous vehicle development by providing scalable, photorealistic simulation capabilities. As the AV industry matures, the ability to generate diverse, realistic test scenarios efficiently becomes increasingly valuable. While practical caveats remain—including potential gaps between simulated and real-world physics—Decart's breakthrough represents meaningful progress toward more accessible and comprehensive autonomous vehicle validation infrastructure. This democratization of advanced simulation tools could accelerate innovation across the autonomous vehicle ecosystem.